JevShield uses typed decisions to block prompt injections locally
A new open-source tool uses Ollama and Nimble to detect prompt injections with structured data instead of generative text, keeping security checks local.
Daily coverage of AI, developer tools and infrastructure. Each story explains what happened and why it matters, with a link to the original source.
A new open-source tool uses Ollama and Nimble to detect prompt injections with structured data instead of generative text, keeping security checks local.
DeepGEMM-Ascend brings API-compatible, near-peak matrix multiplication performance to Huawei Ascend 950 hardware, supporting FP4, FP8, and BF16 workloads.
Opus 5.5 introduces autonomous long-running tasks and built-in reasoning. Learn how to adjust prompts, manage subagents, and handle new safety flags for better coding results.
The Warlock threat actor continues exploiting Microsoft SharePoint vulnerabilities to deploy ransomware against organizations in Portuguese- and Spanish-speaking regions.
ZeroLeaks has released Shield Small, a lightweight 118M-parameter classifier that detects prompt injections and jailbreaks locally on CPU.
Speed of light modeling shows that FSDP creates a communication bottleneck on InfiniBand for DeepSeek-V3, making pipeline parallelism the necessary choice for efficient training.
Anthropic has merged its scheduled task agent, Cowork, into the main Claude app for Pro and Max users, competing with OpenAI Dots and Meta Muse in the race for trusted automation.
Google limits Android accessibility API access to verified tools when Advanced Protection is on, blocking a primary malware vector.
Dell released updates for Container Storage Modules to fix six critical vulnerabilities, including two with perfect CVSS scores, that allow unauthenticated remote attackers to gain admin or root acces
OpenAI's new Dot agent platform prioritizes business software integration and local computer access, contrasting with consumer-focused competitors like Meta Muse.
GitLab released urgent fixes for CVE-2026-90970, a critical vulnerability in self-hosted AI Gateways that allows command execution via crafted prompt templates.
Amazon demonstrates how multi-turn RL fine-tunes a Qwen3.6-27B search agent, cutting failure rates from 22% to under 1% while improving retrieval quality.
CISA adds CVE-2026-104286 to its KEV catalog as attackers exploit a critical path traversal flaw in Fortinet FortiMail to write arbitrary files.
Attackers stole $387.5 million from Bitget by exploiting a zero-day vulnerability in external security appliances and using custom malware to bypass withdrawal controls.
LlamaIndex released Extract v2.5, improving document extraction accuracy across all tiers and adding advanced citations for better grounding.